CadreGPT
Plans & Pricing
Sign In Start Free Trial
Zero-Trust Secrets Isolation

AES-256 Envelope Credential Vault

Store, govern, and dispatch enterprise credentials without ever exposing raw API keys or database tokens to LLM prompt contexts, telemetry streams, or unauthorized operators.

Two-Tier Key Hierarchy

A root Key Encryption Key (KEK) wraps individual Data Encryption Keys (DEKs) generated with unique 96-bit cryptographic nonces. Secrets are stored strictly as encrypted ciphertext in PostgreSQL.

Prompt Context Isolation

Credentials are never passed into model prompts, context windows, or reasoning traces. Decrypted secrets are injected exclusively into outbound HTTP headers at the network layer.

Step-Up Verification

Sensitive credential rotations, API key reconfigurations, and emergency kill-switch toggles enforce mandatory password re-verification to prevent session hijacking.

Zero-Egress Private Vault

Generated artifacts, execution traces, and PDFs are archived directly to private Cloudflare R2 storage with zero egress fees and time-limited pre-signed download tokens.

Ephemeral In-Memory Lifecycle

Decrypted credentials exist strictly within process memory for the microsecond duration of the tool HTTP request. Memory buffers are explicitly zeroed out upon socket completion.

SOC 2 & HIPAA Alignment

Immutable audit logging captures every decryption event, tool dispatch timestamp, and operator identity, providing compliance officers with verifiable audit trails.

Knowledge Base & FAQs

Frequently Asked Questions: AES-256 Envelope Vault

Technical details on cryptographic primitives, prompt isolation, and enterprise key lifecycle management.

01.How does AES-256-GCM envelope encryption differ from standard database encryption?

Standard database encryption (Transparent Data Encryption or disk encryption) protects data at rest if a physical disk is stolen, but leaves secrets in plaintext if the database query engine is compromised. AES-256-GCM envelope encryption utilizes a dual-tier key hierarchy: a root Key Encryption Key (KEK) wraps individual Data Encryption Keys (DEKs) generated with unique cryptographic nonces per secret. Secrets remain encrypted ciphertext inside the database and are decrypted strictly in-memory during authorized tool dispatch.

02.Can an AI agent leak API keys or database credentials in its generated output?

No. Raw integration credentials and provider API keys are never interpolated into prompt contexts or messages sent to LLMs. During tool dispatch, sandbox workers inject authenticated credentials directly into outbound HTTP authorization headers or database connections, completely isolating raw secrets from LLM attention heads and generated outputs.

03.What is Step-Up Password Confirmation for vault credentials?

To protect against unauthorized credential mutation or session hijacking, any attempt to rotate, reveal, or reconfigure sensitive Super-Admin vault keys requires a step-up password confirmation challenge. This ensures that even authenticated active sessions cannot silently alter foundational credentials without re-proving primary authentication.

04.Where are cryptographic encryption keys stored?

The root Key Encryption Key (KEK) is managed via isolated environment secrets and hardware security modules (HSM) / KMS. Ciphertext payloads in the database store the encrypted DEK, IV, and auth tag, ensuring that zero plain-text secrets exist on disk or in database backups.

05.Can our organization bring our own custom LLM endpoints or VPC proxies?

Yes. Super-Admins can configure self-hosted vLLM, Ollama, or private VPC proxy endpoints directly in the vault. All routing tokens and authorization certificates are encrypted under the same AES-256-GCM envelope protocol.

Secure Your Enterprise AI Workforce

Discover how CadreGPT eliminates prompt credential leaks with cryptographic envelope architecture.